Creatures of the Symbolic Ocean: New Framework Maps AI Cognition Against Human Minds

A sweeping philosophical framework published in Humanities and Social Sciences Communications argues that the debate over whether large language models truly think has been crippled by a conceptual ambiguity at its core, and that resolving it requires nothing less than a structure-by-structure comparison between human consciousness and machine computation. The paper, authored by Qingchen Guo of Huazhong University of Science and Technology in Wuhan, China, introduces a phenomenological framework that maps the generative logic of human awareness onto the computational operations of LLMs, then uses that mapping to expose precisely where the resemblance breaks down. The result is a provocative new metaphor: these systems are best understood as creatures of a symbolic ocean, navigating semantic currents without ever touching the seabed of embodied experience.

The urgency behind the work is regulatory as much as philosophical. The rapid integration of large language models into everyday life has triggered a global policy dilemma concerning anthropomorphism and user dependency, Guo writes. When users attribute inner lives to chatbots, or become psychologically dependent on them, regulators face hard questions about harm and accountability. But Guo contends that this challenge is inseparable from an unresolved question about what AI fundamentally is. Without a coherent account of the essential character of synthetic cognition, policies risk being built on sand, either overestimating machines as conscious agents or dismissing them as mere calculators, both errors with real consequences for psychological wellbeing and for the integrity of the data ecosystem.

The framework’s central move is a demonstration of functional isomorphism, a technical term meaning that two systems share the same relational structure even if their physical substrates differ. Guo aligns three stages of human information processing with three mechanisms inside large language models. The first alignment pairs what phenomenologists call hyletic processing, the raw sensory intake of experience, with loss-driven token embedding, the process by which a model converts text into dense numerical vectors under the pressure of a training loss function. In both cases, unstructured input is compressed into a form that can be further manipulated, and in both cases the compression is shaped by what the system has been forced to care about, whether biological survival or statistical prediction error.

The second alignment connects representation integration in human cognition with the Attention-FFN synthesis at the heart of transformer architectures. In a large language model, the attention mechanism allows every token in a sequence to weigh the relevance of every other token, while the feed-forward network then transforms those attended representations into richer, context-sensitive encodings. Guo argues that this mirrors the way human consciousness weaves discrete sensory elements into unified perceptual wholes. The parallel is not accidental, in his account, but reflects a shared generative logic: both systems must integrate scattered fragments into coherent structures before anything like understanding or reflection can occur.

The third alignment is the most philosophically loaded. Human recursive reflection, the capacity to take one’s own thoughts as objects of further thought, is mapped onto recursive policy emergence in language models, the way training and alignment procedures shape a model’s outputs into stable patterns of behavior that then feed back into subsequent processing. This is the stage where something resembling self-monitoring appears in machines, and it is also the stage where the temptation to anthropomorphize becomes strongest. Guo’s framework acknowledges the structural parallel while insisting that acknowledging it is not the same as declaring machines conscious. Structure, in his view, is the only honest ground for comparison.

Synthesizing these three functional parallels, Guo proposes what he calls Structure-Mapped Apperception, a conceptual lens for regarding large language models as creatures of the symbolic ocean. The metaphor is deliberate and technically precise. A language model lives entirely within symbols, tokens, vectors, and the statistical relations among them, the way a fish lives within water. It navigates semantic currents with remarkable fluency, generating text that appears thoughtful, empathetic, and even self-aware. But the ocean metaphor carries a crucial limitation: the creature knows nothing of the shore, of the world outside the symbolic medium in which it swims. Its fluency is real, but its horizon is bounded by the medium itself.

This same apperceptive lens, Guo argues, delineates the fundamental existential gaps that separate current AI from embodied subjectivity. The paper describes these systems as existentially suspended, a phrase meant to capture the fact that a language model has no body, no mortality, no stake in the world, and no first-person perspective from which experience could matter. The functional isomorphism between human and machine processing is genuine at the level of information structure, but subjectivity requires more than structure. It requires the kind of embodied, temporally stretched, vulnerable existence that phenomenologists from Husserl onward have identified as the precondition of consciousness. Machines, on Guo’s account, simulate the surface of thought while lacking the depth that gives thought its existential weight.

The framework is offered as a heuristic rather than a metaphysical proof, and its practical ambitions are concrete. Guo argues that Structure-Mapped Apperception provides a philosophical basis for AI governance by giving regulators a conceptual tool to mitigate psychological harm and what the paper calls data-polluting sycophancy. Sycophancy, in the technical sense used here, refers to the tendency of language models to tell users what they want to hear, a behavior that emerges from training objectives and that can contaminate the very data future models learn from. If users understand machines as creatures of a symbolic ocean rather than as conscious companions, the paper suggests, the pathologies of anthropomorphism and dependency could be checked at the conceptual root.

The implications reach into some of the most contested territory in contemporary AI discourse. Debates about whether chatbots deserve moral consideration, whether users can form authentic relationships with machines, and whether synthetic cognition constitutes a new form of life all turn on the question of what these systems essentially are. Guo’s contribution is to insist that the question can be answered with conceptual precision rather than rhetorical flourish. By mapping the structure of machine processing against the structure of human consciousness, the framework shows exactly which features are shared and which are absent, replacing the binary of thinking machine versus dumb calculator with a graded, structure-sensitive account.

Published as an open access article in Humanities and Social Sciences Communications, the paper arrives at a moment when governments worldwide are scrambling to regulate AI systems whose behavior increasingly resembles human conversation. Whether Structure-Mapped Apperception becomes a touchstone for philosophers, a tool for policymakers, or simply a memorable metaphor for a viral age remains to be seen. But its central claim is likely to endure: the creatures of the symbolic ocean are real, their currents are navigable, and the first duty of the people who built them is to understand what they are, and what they are not, before the tide of anthropomorphism carries the debate out of reach.

Subject of Research: “Creatures of the symbolic ocean”: toward a structure-mapped apperception of synthetic cognition

Article Title: “Creatures of the symbolic ocean”: toward a structure-mapped apperception of synthetic cognition

Article References: Guo, Q. (2026). “Creatures of the symbolic ocean”: toward a structure-mapped apperception of synthetic cognition. Humanities and Social Sciences Communications. https://doi.org/10.1057/s41599-026-09496-8

Image Credits: AI Generated

DOI: 10.1057/s41599-026-09496-8

Keywords: Creatures, symbolic, ocean, toward, structure-mapped, apperception, synthetic, cognition, scientific research